GPU Cloud Provider
TensorWave
TensorWave is a GPU cloud provider focused on delivering high-performance AMD Instinct MI300X GPU infrastructure at competitive pricing for AI training and inference workloads. The company aims to provide an alternative to NVIDIA-centric cloud providers by specializing in AMD's latest data center GPUs.
GPUs
1
Countries
1
Data Centers
1
Team Size
11-50
GPU Marketplace
AMD Instinct MI300X MI300XOn-Demand
Company Profile
Company TypeStartup
Provider TypeBaremetal Provider
HeadquartersUnited States
Legal EntityTensorWave, Inc.
FundingSeries A/B/C/D
Team Size11-50
Infrastructure
GPU FleetAMD Instinct MI300X
Data Center TierCarrier-neutral colocation
Bare MetalYes
EnterpriseResearchStartup
Compute & Deployment
On-DemandYes
Spot / InterruptibleNo
Reserved InstancesYes (monthly and longer-term commitments available)
Bare MetalYes
VM-BasedNo
Container-BasedYes (Docker)
Serverless GPUNo
GPU Hardware
Latest GenAMD Instinct MI300X
Multi-GPU NodesYes (up to 8x per node)
Max GPUs/Node8
NVLinkNo (AMD platform; uses AMD Infinity Fabric interconnect)
InfiniBandYes (400Gbps InfiniBand)
PCIe vs SXMNot applicable (AMD MI300X OAM form factor)
HGX PlatformNo (AMD platform, not NVIDIA HGX)
Pricing Model
Per HourYes (primary billing unit)
Spot DiscountNo spot pricing
Public PricingPartial (some GPUs listed)
Pay-as-you-goYes
Performance & Scaling
Multi-Node TrainingYes (multi-node distributed training supported via ROCm and RCCL)
InfiniBandYes (InfiniBand networking available for inter-node communication)
NVSwitchNo (AMD MI300X uses Infinity Fabric for intra-node GPU communication, not NVSwitch)
Perf IsolationYes (dedicated bare metal instances)
Noisy NeighborYes (bare metal, no sharing)
Developer Experience
OnboardingSelf-service signup with web portal; deploy within minutes
CLI ToolingBasic CLI for instance management
JupyterVia SSH port forwarding
DocumentationBasic getting-started guide with API reference
Data Center Locations
Coverage
CountriesUnited States
CitiesPhoenix AZ
North America
Compliance Regions
EU Data ResidencyNo EU presence
US Gov CloudNo
India RegionNo
Datacenter Locations
Key Strengths
AMD MI300X specialist — 192GB HBM3 per GPU
Alternative to NVIDIA-only cloud providers
Competitive pricing relative to H100 offerings
High memory bandwidth for LLM inference workloads
Bare metal access with no virtualization overhead
Known Limitations
Limited to AMD GPU ecosystem — no NVIDIA options
Smaller fleet size compared to hyperscalers
Limited geographic regions
Ecosystem tooling for AMD ROCm less mature than CUDA
Sparse public documentation and community resources
Additional Information
Core Proposition
AMD Instinct MI300X-focused bare-metal GPU cloud offering high-memory density for large model inference and training workloads.
Payment Methods
Credit CardWire Transfer
Last updated March 2026. Information subject to change.